API: use `__array_wrap__` to instead of `.view(MatrixExpr)`

[Zeroto521]

Is your feature request related to a problem? Please describe.

MatrixVariable is the subclass of MatrixExpr.
The calculation result of MatrixVariable should be MatrixExpr, not MatrixVariable. like #1117

Describe the solution you'd like

NumPy has a protocol __array_wrap__ to fix return type problem.

Additional context

A demo to show that

import numpy as np


class MatrixBase(np.ndarray):
    def __array_wrap__(self, array, context=None, return_scalar=False):
        res = super().__array_wrap__(array, context, return_scalar)
        if return_scalar and isinstance(res, np.ndarray) and res.ndim == 0:
            return res.item()
        elif isinstance(res, np.ndarray):
            if context is not None and context[0] in (
                np.less_equal,
                np.greater_equal,
                np.equal,
            ):
                return res.view(MatrixExprCons)
            return res.view(MatrixExpr)
        return res


class MatrixExpr(MatrixBase): ...


class MatrixVariable(MatrixExpr): ...


class MatrixExprCons(np.ndarray): ...


if __name__ == "__main__":
    a = np.arange(12).reshape((3, 4)).view(MatrixVariable)
    print("Original MatrixExpr:")
    print(type(a))
    print(a)
    # Original MatrixExpr:
    # <class '__main__.MatrixVariable'>
    # [[ 0  1  2  3]
    #  [ 4  5  6  7]
    #  [ 8  9 10 11]]

    b = a + 10
    print("\nAfter addition:")
    # it still is MatrixExpr not MatrixVariable, without changing to .add's default behavior
    # because of __array_ufunc__
    print(type(b))
    print(b)
    # After addition:
    # <class '__main__.MatrixExpr'>
    # [[10 11 12 13]
    #  [14 15 16 17]
    #  [18 19 20 21]]

    c = a.sum(axis=1)
    print("\nAfter sum:")
    # it is still MatrixExpr, without changing to .sum's default behavior
    print(type(c))
    print(c)
    # After sum:
    # <class '__main__.MatrixExpr'>
    # [ 6 22 38]

    d = a == b
    print("\nAfter comparison (a == b):")
    # it is MatrixExprCons because of __array_wrap__
    print(type(d))
    print(d)
    # After comparison (a == b):
    # <class '__main__.MatrixExprCons'>
    # [[False False False False]
    #  [False False False False]
    #  [False False False False]]